CRAIG R. MOORE and JOHN E. PENN

CUSTOM DESIGN OF MONOLITHIC MICROWAVE INTEGRATED CIRCUITS

Monolithic microwave integrated circuits (MMIC'S) have been designed at the Applied Physics Laboratory and fabricated at several gallium arsenide foundries since 1989. The design tools and methods for designing MMIC'S have evolved to the present use of integrated computer-aided engineering software with programmable design components. Software elements that can be customized create multilayer mask descriptions of components for transistors, resistors, capacitors, inductors, microstrip connections, and other structures to improve the quality and productivity of MMIC's designed at the Laboratory. The schematic, physical layout, and simulation models are integrated into a single software tool, eliminating much potential for error. Experience with various foundries and various MMIC design techniques have increased our ability to design at higher and higher frequencies with confidence in achieving first-time success. The design improvements have been accompanied by improvements in measurement techniques for higher frequencies using microwave probe stations. This article summarizes MMIC designs at the Laboratory over the past few years and the progress shown and lessons learned.

COMPUTER-AIDED ENGINEERING DESIGN OF MMIC'S Software computer-aided engineering tools from and custom electrical models. Today, EEsof has a new EEsof, Inc., are the predominant means of designing TriQuint Smart Library containing physical and electrical monolithic microwave integrated circuits (MMIC'S) at APL. macros, which is being used in the MMIC design course A key feature of EEsof's Academy software for MMIC at JHU taught by both authors of this article. layout is the ease of adding both layout macros and A group of people including Dale Dawson of Westing­ custom electrical models. Another feature is the ability to house teamed together to teach a power MMIC design use a particular fixed or macro layout with any available course at JHU that uses a custom physical and electrical electrical model, as long as they have the same number library for Westinghouse's power GaAs process. Employ­ of nodal connections. ees of APL have used Westinghouse's library through the JHU course and for MMIC design on APL programs. In 1992, Macros several power amplifiers were designed and fabricated A macro is a software subroutine that creates the using the Westinghouse library and foundry process. physical mask-layer descriptions for an element that can Some were 2-W amplifiers at 6 GHz, and some were contain variable parameters (Fig. 1). The mask-layer O.S-W designs at 13 GHz. Another 200- to 2S0-mW descriptions are used to generate photo masks for each step in the foundries' integrated circuit processing. An example of a useful macro is a metal semiconductor field­ A B Resistor Metal 1 effect transistor (MESFET), which consists of many mask Metal 2 layers but can usually be described as a given MESFET type with N parallel gate fingers each of width w. One of the first macro libraries used with the Academy \ software was for the TriQuint foundry. It originated at EEsof but was modified by one of the authors of this article, John E. Penn, to create components having no design rule violations. Its first use was in a graduate MMIC Layers(s) defining (JHU) I. .1 ~ design course at The Johns Hopkins University in Length Metal 1 capacitor area the summer of 1989. Since those initial student designs, the TriQuint library was modified by Penn; a graduate Resistance = (length/width). RPA Capacitance = area· CPA student at JHU later provided additional physical macros Figure 1. Example of typical macro elements for monolithic micro­ and custom electrical models. By late 1990, an entire wave integrated circuits. A. Thin-film resistor. RPA = resistance/ multichip wafer was designed by using physical macros area. B. Capacitor. CPA = capacitance/area.

300 Johns Hopkins APL Technical Digest, Volume 14, Number 4 (1993) amplifier at 30 GHz was designed (but has not yet been Plotting fabricated) using Westinghouse high-electron-mobility Calma is the most widely used standard for integrated transistors (HEMT'S). circuit mask descriptions. Using this format, one can The Applied Physics Laboratory created a third MMIC easily transfer integrated circuit designs between tools for library for the GaAs pseudomorphic HEMT (PHEMT) pro­ additional modifications or design checking. This feature cess used by Martin Marietta Laboratories. A fixed set of has been advantageous for obtaining large color plots PHEMT transistors and diodes was used with programma­ from a 36-in. Versatec plotter on the computer-aided en­ ble macros for resistors, capacitors, and microstrip ele­ gineering network. Calma descriptions of MMIC'S are ments. We used the library to design several low-noise transferred to the Mentor Graphics network and then amplifiers and some Schottky diode mixers at 37 GHz for translated into the Chip graph integrated circuit layout a radiometer design for the Geosat advanced technology software for plotting. Large color plots are very useful for model. The library was also used to design a 200- to 250- locating subtle design flaws through visual verification m W amplifier at 29 to 33 GHz for the ultra-smaIl-aper­ and for improving layouts. ture terminal (USAT) program. Measurement Linear and Nonlinear Simulation An essential part of design is measurement of the The simulation engine in EEsof's software is known fabricated MMIC' S to verify that the devices operate cor­ as Libra and is accessible within the Academy frame­ rectly and that the simulation models and design meth­ work. Both linear and nonlinear simulations can be per­ odology are correct. Extra effort to isolate differences formed within Libra with some minor constraints to the between the simulation and measurements will ensure simulation file. Simulations in the linear region generally that future designs can be improved to guarantee first­ use transistors as the active elements where one is con­ pass success. Differences can result from variations in cerned with small signal gain, phase, impedance match, device processing, which can be simulated statistically or and so on. Amplifiers, phase shifters, switching devices, by using best-/worst-case values before sending the de­ couplers, and the like undergo linear analysis, whereas signs to the foundry for fabrication. Measurements of nonlinear analysis is used for RF power, transients, har­ MMIC components (e.g., transistors, capacitors) for resim­ monic response, power efficiency, mixers, oscillators, ulation of the fabricated designs can determine whether transistor current-vs.-voltage curves, and the like. A high­ differences are due to processing changes. Limitations in level simulation at the system level could be considered certain simulation models or the use of models beyond as a third level of simulation where both linear and non­ their specified range can also explain some differences. linear devices are simulated at a basic building-block Omitting parasitic capacitances or coupling from ele­ level. An example of a system-level simulator is EEsof's ments spaced closely on the MMIC should be explored, Omnisys program, which can use data derived from linear or nonlinear simulations or measurements. Ideally, a single integrated simulation tool would perform linear, nonlinear, or system simulations. Microwave ~------J measurement 1------, equipment Electromagnetic Analysis Another useful analytical tool is the high-frequency structure simulator, which is a three-dimensional electro­ magnetic simulator. Reference 1 describes this tool using Needle design examples. * probe Design-Rule Checking

Design-rule checking is needed for any type of inte­ Wafer grated circuit design, including MMIC' S. The DRACULA L----J probe program is the industry standard for design-rule checking MMIC APL. and has been used for verifying designs at Die (chip) Design rules were written by using DRACULA'S program­ ming language and are based on TriQuint's guidelines for the process. In recent years, design-rule checking has been added to the relatively inexpensive pc-based Inte­ grated Circuit Editor layout tool, which can check rela­ /water tively small designs such as most MMIC'S. Design rules for TriQuint's process were created for the Integrated Circuit Editor and tested on APL designs and on student designs at JHU.

*Further description and examples will be given in an article by Jablon, Moore, and Penn to be published in the next Digest, which Figure 2. Wafer probing of monolithic microwave integrated cir­ continues the advanced microwave technology theme. cuits. f ohns Hopkins APL Technical Digest, Vo lume 14, Number 4 (1993) 301 C. R. Moore and J. E. Penn

although parasitics have been found to cause only minor ous calibration schemes to understand measurement ac­ effects on several past APL MMIC designs up to 37 GHz. curacies and the benefits and limitations of the different Being able to resimulate a design to obtain a correct calibration techniques. On-chip and off-chip wafer cal­ simulation file that correlates with the measurement is an ibrations have been compared and shown good agreement important learning experience that ensures future success up to 26 GHZ. 2 Off-chip standards are known high-qual­ and yields confidence in the design tools and methods. ity impedance standards usually purchased from a micro­ MMIc 's are fabricated in large quantities on thin cir­ wave probe manufacturer. On-chip wafer standards are cular disks of GaAs called "wafers" before they are diced impedance standards included on the wafer being fabri­ into individual units and assembled in high-frequency cated that can be used to track processing variations when packages. A microwave probe station allows testing and compared with known off-chip standards or to perform verification of MMIC designs on the wafer before the calibrations for probe measurement. Additional measure­ expensive dicing and assembly stages (Fig. 2). Testing ments were attempted up to 50 GHz, but a distinct diver­ reduces costs by allowing only "good" devices to be gence of the various calibration techniques occurred at packaged, but it requires sophisticated equipment and about 30 GHz and could not be isolated. Some additional techniques. Clever measurement tricks can be used such insight was gained into the coplanar probe to microstrip as adding capacitors to the DC bias needle probes of the launches, but it was unclear how much error might have probe station to eliminate low-frequency oscillations been due to measurement equipment, calibration tech­ when probing amplifiers. We used this technique to probe niques, or assumptions about the coplanar-to-microstrip amplifiers up to 50 GHz. The technique has also been launches. Cascade Microtech, the probe station manufac­ used to probe amplifiers that were considered a failure turer, has recently recommended a calibration technique because they oscillated in previous probe measurements for measurements above 26 GHz. If additional funding that did not use the capacitors. For unstable amplifiers becomes available, we hope to explore measurements up where a capacitor at the needle probes for bias is not to 50 GHz.' enough, amplifiers have been mounted on a small piece of metal with chip capacitors wire-bonded to the bias tIn their upcoming Digest article, mentioned in the previous footnote, MMIC pads on the to allow successful probe testing. Jablon et al. will discuss the use of a three-dimensional electromag­ Consistent, precise measurements are as important to netic field solver to provide insight into the on-wafer MMIC calibration designing MMIC'S as the design tools. We have used vari- standards.

·• , ...... •• ·iI .,. ·..,...... ~ . • f •••••••. •• 1. .:• 1::•••••1::1: ••••••••••••••••••• • •• : tf I·· .~~J):~ ~ ,~ .,. )~j: ~;:( .. : .•. ... .•• .. : ~ \ . .. :, 1, ;., ,,; ..0: .. , t ·- : • •• • I •• I • '.: .. ' •• >< I . ~ ---- ~.; . .>< . ••. ~: ~ ..••••••••••••••••••• - ,., . '. " .. "

Figure 3. Multichip wafer reticle of a monolithic microwave integrated circuit tracking receiver. The entire 360 x 360 mil tile was reproduced seventy-six times on a 4-in. GaAs wafer.

302 Johns Hopkins APL Technical Digest, Volume 14, Number 4 (1993) Custom Design of Monolithic Microwave Integrated Circuits

APL MMIC DESIGNS testing of the amplifier, it was initially oscillating and thought to be unstable despite the previous simulations. S-Band MMIC Tracking Receiver Because of the long bias wires and artificially high in­ The MMIC tracking receiver program produced the first ductances created in the test configuration, an amplifier full custom GaAs multi chip microwave wafer designed may oscillate at low frequencies. Stable probing of the at APL. Using the TriQuint macro library described earlier, amplifier was achieved by adding bypass capacitors at the all of the designs were simulated and physically laid out needle probes used to apply DC bias to the device being on pc's using EEsof software. The designs were also tested. checked for violations at APL by using the DRACULA pro­ Westinghouse's power GaAs process was used to cre­ gram. Nineteen unique MMIC' S were fabricated, including ate multiple versions of a 0.5-W power amplifier for the amplifiers, mixers, hybrids, and active filters from 300 microwave remote sensing element program. Some of the MHz to 13.6 GHz. A large MMIC supercomponent was devices included switches to shunt the outputs when the included that contained a complete image-rejection (I and module was in a receive mode, thus performing the trans­ Q) mixer at 2.3 GHz RF and 300 MHz IF. Figure 3 is a mit/receive switch function. All of the fabricated devices plot of the entire 360 X 360 mil tile, which was repro­ were functional. Two power amplifier versions, one with duced seventy-six times on a 4-in. GaAs wafer. a switch and one without, were measured by bonding the devices to a piece of metal and wire-bonding bypass chip K-Band Microwave Remote Sensing Element capacitors to the supply pads to stabilize the devices. A Another multi chip wafer was fabricated for an active­ probe station was used to measure the amplifiers, and DC element array sensor. Amplifiers, phase shifters, attenu­ bias was provided by needle probes touching the tops of ators, mixers, transmit/receive switches, and power am­ the chip capacitors. The losses in the test setup were plifiers were designed to create a transmit/receive module subtracted to determine the large signal power out of the for the active-element phased array. Test structures for devices. Although slightly below the design frequency, measurement were included in unused space across the the measured peak output power curves closely matched wafer. All of the devices except the power amplifier were the shape of the simulations. One power amplifier that fabricated by TriQuint on a single multichip wafer. contained switches and was expected to achieve about One of the designs was a five-stage high-gain amplifier 400 mW of output power (+26 dBm) did so, as shown for transmit and receive. One of the authors of this article, in Figure 5. Another design variation, which had about Craig Moore, designed a simple driver that took in stan­ one-third more output MESFET periphery and no output dard digital control signals that were transistor-transistor switches, was expected to achieve over 630 mW (+28 logic compatible and created appropriate analog voltages dBm). It also met or exceeded the expected output power to turn the amplifier gate bias on or off, depending on with an overall efficiency of about 25 % for the three­ whether the module was transmitting or receiving. A stage amplifier. single transmit/receive digital control signal replaces two Several test structures, such as capacitors, inductors, control signals, since each amplifier is configured for resistors, FET'S, and microstrip structures, were included either the transmit or receive function by breaking one on these wafers for calibration and verification. Measure­ of two air-bridge metal structures. With each amplifier ment and calibration techniques using the impedance modified in this way, the transmit amplifier will be off standard substrate from Cascade Microtech, Inc., and when the receive amplifier is on and vice versa. In the "on-wafer" structures allowed us to measure the devices amplifier state, there is 35 dB of gain (Fig. 4), and in the accurately up to 26.5 GHz. The good agreement of the off state there is over 70 dB of isolation! During probe measurements using various calibration techniques gave

B 40,------,------.------.-----~------~

A 20

o

20 Frequency (GHz)

Figure 4. A. Photograph of a K-band microwave remote sensing element (MRSE) amplifier. B. K-band MRSE amplifier gain in the on (black) and off (blue) states.

Johns Hopkins APL Technical Digest, Vo lume 14, Number 4 (1993) 303 C. R. Moore and J. E. Penn

A A o o '~1 - - ~ -. i-:I ~ r<.!: r<2i • • III [J B 30~------~------~------~----~ 30

B 24 / 24 20 ~----~------~------~------~-----. / E / a:l ~ 18 / Q; 3: / o 10 a. / "5 12 / a. o"5 6 6

OL------L------L------~----~ o -10 -10 -6 -2 2 6 Input power (dBm)

Figure 5. A. Photograph of a typical amplifier. B. Output power -20L-----~------L------~------~----~ (black) and efficiency (blue) of a typical amplifier. The solid curves o 10 20 30 40 50 are the simulated values, and the dashed curves are the measured Frequency (GHz) values. Figure 6. A. Photograph of a radiometer amplifier. B. Simulated gain (black) versus measured gain (blue) for a radiometer ampli­ fier. us a standard with which to compare the simulation models. As a result, some of the foundry-supplied models were improved. Also, the MESFET measurements allowed resimulation of the designs to see whether differences between the original simulations and the measured devic­ es were due to wafer processing variations or to errors 20 8 in the simulation models.2

Ka-Band Radiometer 15 6 U sing a simple macro library for the Martin Marietta Ci) GaAs process, amplifiers and mixers were designed and Ci) ~ -a ~ ::J fabricated for the Geosat Advanced Technology Model -; 10 4 OJ 'co 'i= program. Martin Marietta has a very low-noise, high­ (9 Q) U) frequency PHEMT GaAs process used for designs at 94 '0 z GHz and above. Programmable macros for microstrip 5 2 elements, resistors, capacitors, and spiral inductors were created for their process . The PHEMT'S and diodes were added using a small library of fixed, active-component 0 0 layouts. Several low-noise amplifiers were designed for 25 30 35 40 45 37-GHz operation. One was designed for lowest noise Frequency (GHz) figure, another for moderate noise figure with good input and output match (voltage standing-wave ratio), and a Figure 7. Gain and noise figure of a Ka-band low-noise amplifier. third for high gain with moderate noise figure. Also, Simulated gain (black), simulated noise figure (blue), and mea­ several variations in the bias supply methods were tried sured gain using an HP 8510 probe (red). The circles refer to on a single amplifier design to see which technique measured noise figure and gain using an HP 8970 probe. The vertical ranges around the circles indicate the scatter of the worked best at 37 GHz. Another new concept used in measurements, and the horizontal range centered at 37 GHz is the these amplifier designs was "wafer tuning," introduced to frequency range of the measurements.

304 Johns Hopkins APL Technical Digest, Vo lume 14, Number 4 (1993) Custom Design of Monolithic Microwave Integrated Circuits us by Martin Marietta; it provides a means of tuning the branch-line couplers. The mixers should be tested in the MMIC' S. Two microstrip stubs with several breakable air near future. bridges spaced 75 /Lm apart were used to tune each amplifier. Although the tuning and testing of the first Ka-Band Power Amplifier device may be tedious, the other devices on the wafer can A power amplifier was needed for a small, low-cost be expected to operate similarly. One need only tune the USAT. A 0.25-W amplifier at 30 GHz was designed using first device or two, then set all the devices from that wafer MMIC 'S. A first iteration employed the Westinghouse pro­ to the same tune point for similar operation. Generally, cess for GaAs HEMT'S. About 200 to 250 mW of output the devices operated fairly well without any tuning. Fig­ power was expected at 30 GHz with about 20 dB of gain ure 6 shows the measured and simulated gain versus in the three-stage amplifier. An alternative design used frequency of one of the amplifiers. Note the wide band­ the Martin Marietta PHEMT GaAs process; for this design, width of the 20- to 40-GHz operation. The noise figure several PHEMT'S were measured, and a nonlinear model was measured for some devices and compared favorably was extracted using a probe station, network analyzer, with the simulations (Fig. 7). When testing for noise and Hewlett Packard IC-CAP software. Since the gain of figure, one must adjust the MESFET bias point for lowest the transistors was higher in the Martin Marietta process, noise, since the transistor performance is sensitive to the only two stages were required to achieve about 20 dB of DC operating point. small signal gain. Wilkinson couplers were used as power A diode mixer operating at 37 GHz was designed using dividers/combiners so that maximum gain with good the Martin Marietta Schottky diode GaAs process. The matching could be obtained in the final stage. As in the same macro library was used, but the actual GaAs process previous Westinghouse design, four PHEMT'S were com­ was slightly different for the diode process and the PHEMT bined in the output stage to achieve 200 to 250 m W of process. Several mixer variations were fabricated using power at 29 to 33 GHz. The layout and simulations of hybrid branch-line couplers and Lange couplers. The the Martin Marietta power amplifier, currently in fabri­ Lange couplers were smaller but have higher loss than the cation, are shown in Figure 8.

A

B C 30~----~------~----~------~-----. 0 25 ~------~------~------.------.--, 25

E20 20 iii' en (i) -10 :3- :3- i en >- :3- en Ci5 0 .Q 15 c c ~ 15 Q) 'co c Q. 'u (9 ::; 10 -20 Q5 iD a: 10

5L-______L- ______L-L- ____L- ______L- ~ oL------L------~----~------~----~-30 5 25 27 29 31 33 35 -14 -10 -6 -2 2 Frequency (GHz) Input power (dBm) Figure 8. A. Layout of a 30-GHz amplifier. B. Simulated small-signal S-parameters of a 30-GHz amplifier. Gain (black) , output return loss (blue), and input return loss (red). C. Output power (black) and efficiency (blue) versus input power of a 30-GHz amplifier.

Johns Hopkins APL Technical Digest, Volume 14, Number 4 (1993) 305 C. R. Moore and 1. E. Penn

CONCLUSION REFERENCES The Laboratory has significantly advanced MMIC de­ 1Penn , 1. , and Moore, c., "Model Verification of Passive MMIC Structures Through Measurement and 3D Finite Element Simulation," in Proc. Johns sign, testing, and modeling during the past few years. Hopkins University Microwave Symp., pp. 1-3 to 1-13 (Oct 1992). Future work will continue to push higher frequencies of 2Penn, 1. , and Moore, C., "GaAs MMIC Probe Measurements and Calibration operation that will challenge both the design and testing Techniques," in 39th ARFTG Conf. Digest, pp. 106-119 (Spring 1992). of MMIC'S. The goal is to build on previous work by ACKNOWLEDGMENTS: We would like to acknowledge the contributions of compiling a computer library of component designs that Robert Bokulic, Lynn Carpenter, Sheng Cheng, Charles Cook, Matthew Reinhart, can be used at the system-design level. Jeffrey Sinsky, Roy Sloan, Karl Thompsen, and William Trimble.

THE AUTHORS

CRAIG R. MOORE is a member JOHN E. PENN is a member of of APL 's Principal Staff. He re­ APL'S Senior Staff. He received a ceived B.E.E. and M.S. degrees B.E.E. from The Georgia Institute from Cornell University in 1962 of Technology in 1980, and M.S. and 1964. He previously served as degrees in electrical engineering associate division head at the Na­ and computer science from The tional Radio Astronomy Observa­ Johns Hopkins Unversity in 1982 tory, principal research associate and 1988. Mr. Penn joined APL in for the Australian government, and 1980; his work at the Laboratory group manager at Bendix Field has involved custom integrated cir­ Engineering Corporation, before cuit design, semi-custom design, joining APL in 1987. His work at silicon compiler design, digital de­ the Laboratory has involved im­ sign, programming, and, more re­ provements to the hydrogen maser, cently, microwave design. He left work on ultra-high-Q cryogenic APL in 1990 to work as an applica­ microwave resonators, and MMIC tions engineer for Silicon Compiler design and testing. He is currently Systems, which became part of project manager for an ultra-small satellite terminal design effort. Mr. Mentor Graphics. In 1992, he rejoined the Laboratory in the S2R Moore is a senior member of the IEEE and has authored more than Group, where he has been designing MMIC'S. Mr. Penn co-teaches a twenty papers. He teaches MMIC design at The Johns Hopkins MMIC design course with Craig Moore at The Johns Hopkins University G.W.c. Whiting School of Engineering. University G.W.c. Whiting School of Engineering.

306 Johns Hopkins APL Technical Digest, Volume 14, Number 4 (1993)